the impact of computer-based trading on market quality · 2020. 8. 19. · oliver linton university...

The Impact of Computer-Based Trading onMarket Quality

Oliver LintonUniversity of Cambridge

11th January, 2013

Oliver Linton University of Cambridge

UK Government Offi ce for Science Foresight project on The Futureof Computer Trading on Financial MarketsInternational project with hundreds of academics and practitionersinvolved


Some Questions our project addresses

1 What are the effects of Computer-Based trading ("CBT"including "AT" and "HFT") on market quality measures likeliquidity, transaction costs, price discovery, and effi ciency innormal times?

2 What are the risks associated with the changing nature ofmarket making (which has shifted from designated providersto opportunistic HFT traders with in many cases limitedcapital commitment combined with ultra-fast speed) inextreme times?

3 What are the appropriate definitions of marketabuse/manipulation in a high speed multi-venue multi-assetenvironment? Has the advent of CBT increased marketmanipulation?


Main complaints about HFT (from e.g., OW survey ofbuy-side)

Liquidity they supply is ephemeral. Narrow bid ask spreads,but created by "flickering quotes", i.e., not accessible tohumans. Pass the parcel trading between HFT. Liquiditysupply evaporates during crisis times e.g., Flash Crash.Too much messaging, i.e., order cancellations and revisions,negative externality. Requires big investments in technology(Smart Order Routers, Co-location, Algos for order slicing) tokeep up. Arms race for speed socially unproductive (PaulKrugman)Increase volatilityMarket abuse and manipulation: quote stuffi ng, spoofing,layering, smoking, etc.


Theoretical Arguments on either side

Jarrow and Protter (2011), Biais et al. (2012), Pagnotta andPhilippon (2012), Cartea and Penalva (2011), Jovanovic andMenkveld (2011).

No dominant agreed-on theoretical model. Furthermore, inmany cases, models suffer from oversimplifications either:

No explicit time scale, when time is the key feature here.Simplified strategy space for traders (HFT contains a numberof different trading styles: market making (e.g., GETCO),statistical arbitrage, directional or momentum traders. Thefirst two typically thought to have good contribution to themarkets, the third less so. Same firm can combine allstrategies differently over time)Simplified market structure


Caution about What we Know and What we dont Know

The question of whether it is good or bad can only beanswered with empirical evidence and with reference to somebenchmark alternative universe.

Recent phenomenon of interest. Not much published researchon this. We survey what there is with an emphasis on highquality peer reviewed work.

Field is evolving so some caution is required in makingconclusions. It is not like the issue of whether smoking hasnegative health outcomes where there is almost universalagreement based on lots of careful research. It is also notperhaps at the same level as climate change...

The research on CBT being so recent has not achieved thatlevel of agreement.


System latency following Moores law or even more

System Implementation Date Latency (Microseconds)

SETS <2000 600000SETS1 Nov 2001 250000SETS2 Jan 2003 100000SETS3 Oct 2005 55000TradElect June 18, 2007 15000TradElect 2 October 31, 2007 11000TradElect 3 September 1, 2008 6000TradElect 4 May 2, 2009 5000TradElect 4.1 July 20, 2009 3700TradElect 5 March 20, 2010 3000Millenium February 14, 2011 113


What is the Social Cost of High Frequency Trading

Costs of speed ↓; Benefits of speed are small but not zero. Race tothe bottom. Krugman, Haldane, etc.Can we bound the social cost of the socially unproductive armsrace for speed?What is the value of the Equity Trading Space?Profitability estimated $8b-20b per year in US. Kearns et al studytries to bound profitability of omniscient HFT more like $3b.Compare with total value of electronic order book trading $30tr in2009Knight Capital. US electronic market maker, 10% of total value ofequity traded in US in 2011. Total revenue $1.4b, net income$114m in 2011. 1500 employees.Not trivial, but compare with Exxon Mobile Net Income of$41billion in 2011


What Kind of Evidence?

The evidence we consider is of two typesEvent studies/natural experiments (exogenous changes inconditions that favour HFT but do not affect directly outcomeof interest)Time series comparison of now and then

The main issues are:

Measurement of HFT/AT where this is necessary. Somestudies work with "trader id’s" that classify tradersMeasurement of market quality in high frequency settingEndogeneity/causality issues. For example, HFT may causevolatility and/or volatility may cause increased trading andhence HFT.Time scale to do comparison. Is it meaningful to comparetrading activity within a second between 2012 and 1912?


Rising Tide Lifts all Boats?

The level of the stock market has gone up and down violently overthe new millenium. The macro economy has been in turmoil withdreadful performance since 2007. These are giant factors outthere affecting savers and investors and any analysis that is doneneeds to control for this.


Liquidity

Tightness, Depth, Resilience, but diffi cult to define precisely.Common measures bid ask spreads, effective spreads, realizedspreads, depth, trading volume etc.

Hasbrouck and Saar (2010) Nasdaq during 2007-2008 orderbook. Low latency activity improves spreads and depth.Hendershott, Jones, and Menkveld (2011, Journal ofFinance). How does CBT/HFT activity affect liquidity? Theyuse the phased in automation of the NYSE quote system in2003 as a natural experiment to measure the causal effect ofCBT on liquidity.Their findings were that CBT improves liquidity and enhancesthe informativeness of quotes.Brogaard (2011) dissertation, US data. Several PhD studentson market this year with new work on this data


Castura at al. Show bid ask spreads and depth on the Russell1000 and 2000 over the period 2006-2010. They show theimprovement of bid ask spreads and depth over this time.

Payne driver shows the same for FTSE100 stocks over theperiod Jan 2009 to April 2011 and finds improvements


J&M. Empirical study of the entry of a HFT player into themarket for Euronext listed Dutch index stocks on Chi-X in2007/2008.

Compares with Belgian stocks which had no such treatment.Difference in difference analysis: before and after (it wasduring a busy time for markets), Dutch and Belgian

Shows improved market quality metrics, reduced adverseselection components, and more trading. Chi-X quotes moreresponsive to futures quotes than Euronext. Consistent withthe story that middleman are more active during periods ofhard information

Gresse (2011) shows that the fragmentation of trading(associated with HFT activity) after MiFID1 lead toimprovements of effective and realized spreads as well asmarket depth especially for larger stocks in Europe.


Transaction costs

Angel et al (2010) show that average retail commissions inthe USA have decreased between 2003 and 2010. Impactcosts also less evidenced from bid ask spreads.

Menkveld (2011) investigates the effect of the entry of Chi-Xinto the market for Dutch index stocks. Had an immediateand substantial (negative) effect on trading fees for investorsthrough

The lower fees that Chi-X chargedThe lower fees that Euronext then chargedThe reduction in clearing and settlement fees that followed


Price Effi ciency/Discovery

Hendershott driver reviews the arguments why HFT mayimprove effi ciency by bringing into line prices of correlatedsecurities, buying low and selling high.Main evidence is from Castura et al. study. High FrequencyVariance ratios (They look at 10:1 second variance ratios aswell as 60:10 and 600:60 second ratios) on Russell 1000 and2000 over the period 2006 to 2010 have come closer to one,meaning less linear predictability.Low frequency effi ciency metrics show no trend towardsimprovement or disimprovement. Daily FTSE all share datafrom 2000-2012: variance ratios fluctuate over the businesscycle and there was more evidence of predictability during the2007/2008 period but that lack of effi ciency has declinedsince then.


Rolling window (annual) VR(5) with iid confidence bands


Some evidence that announcement effects are much fasterimpacted into prices. A positive US non-farm payroll day in 2012


Knight Capital. US market maker. Listed on NYSE. August 1,2012 trading error caused widespread disruption on NYSE. Theylost $450m in a few minutes.Their own share price reflected that disaster, over the day andovernight. 20120801 open 10.33 close 6.94, 20120802 open 3.30close 2.58


Volatility

Not necessarily a good or a bad except makes life diffi cult for someinvestors. Financial stability.Some studies show slight increase in volatility due to HFT activity(messaging) - Zhang (2010), Boehmer (2011), and OHara and Ye(2011).Other studies show decreases, Brogaard (2011), Brogaard et al(2012). Hasbrouck and Saar (2010) Low latency activity reducesvolatility. Chaboud et al. (2009) (xrates) CBT associated withslightly lower volatility.


Time series graph of FTSE100 share index daily volatilitymeasure 1984-2012.

High minus low measure of volatility takes account of allvalues that price takes within a day so that extrememovements are not missed. Really an ultra high frequencymeasure.

This graph shows that volatility has gone up and down overthe last twenty years. In 2007 it started picking up andincreased to a peak in 2008, after which it lowered throughout2010.


FTSE100 Top 20 Most Volatile days since 1984 (- meansPC < PO , + means PC > PO )

Date Volatility

19871020 0.131-19871022 0.115-20081010 0.112-19971028 0.096-20081024 0.096-20081006 0.094-20081008 0.094-20080919 0.093+20081124 0.090+20081015 0.084-19871019 0.081-20020920 0.080+

20081013 0.076+20010921 0.076-20081029 0.075+20110809 0.074+20090114 0.074-20080122 0.074+20020715 0.071-20081016 0.070-


S&P500 Top 20 Most Volatile (high-low/close) days since 1960

Date Volatility

19871019 0.257-19871020 0.123+20081010 0.107-20081009 0.106-20081113 0.104+20081028 0.101+20081015 0.100-20081120 0.097-20081013 0.094+20080929 0.093-19871026 0.092-20100506 0.090-

20081201 0.089+19620529 0.089+19871021 0.087+20081016 0.087+20081006 0.085-20081022 0.085-20020724 0.081+19980831 0.080-

US market more volatile than UK. Both sides of Atlanticdominated by 2008 and 1987


Crashes or Market Behaviour in Extreme Times

Many financial time series are subject to large movements, i.e.,have heavy tails relative to the normal distribution. Mandelbrot(1963). The biggest crashes have usually been caused byfundamental events, e.g., the Russian revolution effect on theRussian stock market.Emerging market exchange rates: Russian ruble after default, ThaiBaht during Asian crisis.Electricity prices can be subject to large spikes due to supplyshortfalls, Weron (2008).Gopikrishnan et al. (2000) find that trading volumes for the 1,000largest U.S. stocks have Pareto tail with exponent around 3/2Sornette and his dragon klingons


The Flash crash was relatively small, and not contagioned tooutside world unlike the 1929 and 1987 crashesBut at least part of it was driven by purely technological/tradingstructure failure, which is of concern. Endogenous risk.Kirilenko and Kyle study of futures markets during flash crash.Shows fragility of liquidity provision in the presence of order flowtoxicity (Easley and OHara)Our study did not focus on flash crash and its causes - nocomparable event (in impact) in European equity markets andmarket structure here differentBut it could happen here and we comment later on some policymeasures designed to mitigate risks.


Concluding Remarks

1 A lot of negative articles and comments about HFT. Despitethis, the best evidence we have today suggests that CBT ingeneral and HFT in particular have several beneficial effectson routine market quality

1 They have contributed to improvements in the Liquidity ofmarkets as measured by bid-ask spreads and other metrics

2 They have contributed to improvements in Transaction costsboth for retail investors and for institutional investors mostlydue to changes in market structure which are related to thedevelopments of HFT

3 They have contributed to improvements in Market effi ciencyby impacting new information into prices faster and by linkingfragmented market places together


1 However, while liquidity has improved overall, there appears tobe the potential for increased periodic illiquidity or liquiditycrises such as the US flash crash of May 6th 2010 and variousother smaller events since then (predominantly in the US)

2 This can arise through feedback loops generated within thecomputerized trading process itself that once started canamplify over time even within well intention management andcontrol processes


1 There is no direct evidence that computer based trading hasincreased market abuse

2 However, defining and detecting market abuse in todaysenvironment requires more and better data than is currentlyavailable

3 The perception of abuse itself may harm market qualitythrough investor withdrawal and this needs to be countered byregulators demonstrating a firm commitment and ability tocounter market abuse in todays fragmented and high speedtrading environment


the impact of computer-based trading on market quality · 2020. 8. 19. · oliver linton university...

Documents